One Union, two opposite screens: Huawei versus the hyperscalers¶

A plain-language companion to the working paper "Two Screenings: Why Europe Screened Huawei but not the Hyperscalers" (v0.5, July 2026). The paper carries the primary documents and the analytical frame; this text carries the ideas.

The short version. Within one decade the European Union faced the same problem twice: suppliers of critical digital infrastructure whose home state can compel them. Against Chinese telecom vendors it wrote a structural screen: assess who owns and controls the supplier, and restrict accordingly (the 5G Toolbox, January 2020). Against US cloud providers it did the opposite: the structural criteria drafted into the EUCS certification scheme were deleted in March 2024, leaving behavioural requirements about residency and operations. Same polity, same problem, opposite designs. The explanation needs exactly two variables, and it makes testable predictions that events since have kept confirming.

The two variables¶

How verifiable is the decisive attribute? For a Chinese state-linked vendor, state control is presumed and public: no audit needed, no subsidiary structure hides it. A screen on it is administratively trivial, and any parliament can read it. For a US hyperscaler, effective control behind the EU subsidiaries must be established, against sophisticated legal engineering, which makes the criterion expensive to verify and easy to argue about in a drafting room.

Where can the losing coalition spend its influence? Influence flows to wherever discretion is highest and salience lowest. A legible, security-framed rule is expensive to attack in daylight: weakening it visibly reads as siding with the threat. A hard-to-verify criterion inside a technical certification scheme can be deleted quietly, and the deletion reads as drafting.

The record, as a matrix¶

Against the 5G screen stood a coalition of customers: telecom operators facing large but one-off, compensable replacement costs. It lost at the design stage (the Toolbox passed with its origin criteria intact) and won quietly at enforcement: three and a half years in, 24 of 27 member states had legal powers and only 10 had imposed restrictions, with national outcomes tracking domestic exposure. Against the cloud screen stood the suppliers themselves plus their ecosystem, with permanent stakes measured in ecosystem rents. It won at the design stage: the criteria died in a low-salience venue before there was anything to enforce. Four cells, one regularity: capture goes where the rule is cheapest to bend.

The predictions, each with a way to fail¶

The frame was, in effect, calibrated on events through early 2024, so what happened next is evidence. It predicted venue migration: structural criteria, defeated in a capturable certification venue, should reappear where discretion is inherent and salience higher: procurement and legislation. That happened: the Commission's Cloud Sovereignty Framework (October 2025) ranks sovereignty as a procurement dimension, a ~EUR 180M tender applied it within months, and the Cloud and AI Development Act proposal (June 2026) moves ownership-and-control tiers into law. It predicted enforcement hardening where capture had won at enforcement: the January 2026 Cybersecurity Act revision proposes making 5G derisking binding. And it predicted that within-EU variance would track domestic exposure, which the cross-country record supports. Each prediction carries a live falsifier, stated in the paper: if CADA's structural tiers die in trilogue, or the Cybersecurity Act revision dies quietly in Council, the explanation loses.

What the explanation is not¶

It is not a claim that alliance politics plays no role. Cross-country evidence makes security dependence on the United States the best single predictor of Huawei bans, and the paper concedes the level difference between the two cases to it without resistance. But alliance position is constant within the EU and across the decade, so it cannot explain the within-EU enforcement variance, the venue migration, or the timing. The two-variable frame can, and it is the part that generalizes: wherever a polity screens a supplier whose decisive attribute is hard to verify, expect the fight to happen in the drafting room, and expect the criteria to die there quietly.